Two-stage ensemble detection
A routing model sends your X-ray to the optimal specialist — panoramic, bitewing, or periapical — then two independent neural networks cross-reference findings and come to a consensus.
YOLOV8 · YOLOV11WBF · IoU 0.6
KIRANA/SCAN · PANORAMICSTANDBY
INFERENCE0%
LOADSCANDETECTREPORT
PT · 4829-LW· PANO · 66kVp · 10mA · 13.5s
FINDINGS0/8
Q2UPPER LEFT · 21–28
Q1UPPER RIGHT · 11–18
Q3LOWER LEFT · 31–38
Q4LOWER RIGHT · 41–48
28Impacted
0.00confidence
38Impacted
0.00confidence
36Caries
0.00confidence
37RCT
0.00confidence
36RCT
0.00confidence
35Periapical
0.00confidence
46RCT
0.00confidence
48Impacted
0.00confidence
Clinical dental AI · CE-ready
See every lesion
with a second
pair of X-ray eyes.
Kirana Scan runs an ensemble of specialist models across panoramic and bitewing X-rays — every finding carries a calibrated confidence score and a structured, IR(ME)R-compliant report in under three seconds.
THE THESIS
A single model is a single opinion.
Clinical accuracy demands an ensemble.
01 / DETECT
Four specialist models, one clinical answer
We run image-type routing, tooth-identification, and X-ray type specific models.
No single model decides alone.
02 / REPORT
Structured narratives, not raw boxes
A vision-language model turns the detection graph into an IR(ME)R-compliant report with ICCMS grading and treatment notes, ready for your letterhead in under 45 seconds.
03 / AUDIT
Every inference is clinically traceable
Per-finding confidence, training-cohort provenance, and out-of-sample validation on thousands of unseen X-rays. Nothing is a black box for your medico-legal file.
CAPABILITIES
Clinical-grade detection, explained.
A live scan of a panoramic radiograph — rendered directly from Kirana Scan's detection graph. Findings are ranked by calibrated confidence and routed to the correct specialist model.
X-ray-specific specialist models
Kirana Scan identifies panoramic vs bitewing vs periapical automatically, then routes the image to the model tuned for that modality — no manual selection, no misapplied weights.
ROUTING · 99.1%3 MODALITIES
IR(ME)R-ready clinical reports
Vision-language model generates structured reports with ICCMS grading, findings, and treatment notes for every patient.
MEDGEMMA-27B
Validation on 10,000s of unseen X-rays
82% sensitivity, calibrated confidence, and per-class thresholds tuned with Bayesian optimization against held-out clinical data.
OOS · 13,042
Under 10s end-to-end
Routing, ensemble detection, tooth charting, and the clinical narrative — all complete before your patient is back in the chair.
P50 · 42s
FDI tooth notation, automatic
Every finding is anchored to a specific FDI-numbered tooth, with missing and impacted teeth charted for you.
32 CLASSES · 91% ACC
20,000+
Training images
Curated from six open and partner clinical datasets, annotated to ICCMS standard.
82%
Ensemble sensitivity
Out-of-sample on 13,042 unseen X-rays — catching what single models miss.
4
Pathology classes
Caries, periapical lesions, impacted teeth, and root-canal treatments.
<10s
Report turnaround
Upload to signed, IR(ME)R-compliant PDF. P50 latency 42s on standard hardware.
PIPELINE
Four stages, one clinical answer.
Each stage is an independent, specialist model. Failures in one don't silently corrupt the others — the pipeline is auditable at every step.
STAGE 01PANORAMIC
Panoramic detection
Two independent networks analyse your panoramic X-ray and fuse their results. Catches what a single model misses.
SENSITIVITY82%
CariesPeriapicalImpactedRCT
STAGE 02CHARTING
Tooth identification
Automatically detects and numbers every visible tooth using FDI notation. Builds a complete dental chart.
ACCURACY91%
32 FDI classesMissing teeth
STAGE 03BITEWING
Bitewing analysis
A dedicated model trained for interproximal caries on bitewings. Activates automatically when the routing model detects a bitewing image.
F2 SCORE0.78
InterproximalBitewing-tuned
STAGE 04REPORT
Clinical reporting
A vision-language model generates an IR(ME)R-compliant report with ICCMS grading, structured findings, and recommendations.
COMPLIANTIR(ME)R
ICCMSStructuredEditable
CLINICIANS
What dentists actually say after two months.
“
Kirana Scan flagged a periapical lesion on a lower molar that I'd initially overlooked during a packed Monday morning clinic. That early catch made all the difference for the patient's treatment plan — and honestly, for my confidence on a tired day.
SM
Dr. Sarah Mitchell
Lead Dentist · St Vincent Smile
“
Report writing used to eat into our appointment slots. Now the AI drafts a structured report in seconds and our dentists just review and sign off — we're seeing three more patients a day.
JO
James O'Brien
Practice Manager · St Vincent Smile
“
I use it as a second pair of eyes, especially for subtle interproximal caries. It's reassuring when the AI confirms what I'm seeing — and it's taught me to look more carefully when it catches something I missed.
PS
Dr. Priya Sharma
Associate Dentist · St Vincent Smile
START A 30-DAY PILOT
Put an ensemble of specialist eyes on every X-ray you take this afternoon.
No training required. Upload a panoramic or bitewing, get back an IR(ME)R-compliant report before your next patient is seated. Free for the first 100 scans.